Project objectives.
Arobas Technologies was looking to develop its HPC resources around two main elements.
The idea was to add a hyper-frequency node for application with mainly sequential calculations, but also for when licencing costs become too high so that all the new processor’s cores are available.
The second aspect was to upgrade the dispersed server (parallel file system) for scratch space (temporary write files) to ensure there are no restrictions to parallel calculations.
The starting capacities of around 20 TB needed to be easily scalable, simple to manage and able to be integrated into current internal resources with the help of the Bright Cluster Manager.
BeeGFS has been integrated into Bright since 2017 and can deliver extremely high performance when configured correctly.
The comprehensive update to HPC infrastructure was very efficient, with the new version of Bright (8) and BeeGFS being installed on our cluster in only three days. It was a critical operation because our resources were unavailable during this period, but everything went to plan and the temporary suspension of our services was completely under control from start to finish.
Jacky Verguet, CEO, Arobas Technologies
Solution.
Fra-SYS put together an overclocked machine based on an Intel i9 CPU that enabled Arobas Technologies to use up to 18 overclocked compute cores with nearly 4.2 GHz. The performance increase can be over 40% for some applications compared to servers with two processors.
A new BeeGFS storage solution was introduced together with an update to the Bright Cluster Manager. The hardware was specially selected from Supermicro’s catalogue—a dense enclosure (2 U) with 12 slots for capacitive hard drives (9 x HDD 2 TB in RAID0, net capacity of about 16.7 TB) and/or very fast hard disks (3 x SSD 240 GB in RAID5), but also two dedicated slots for 2 x SSD (2 x SSD 240 GB in RAID1 for the operating system). These storage nodes also have a 128 GB capacity and an 8-core CPU. The connection to the existing servers is via a dedicated InfiniBand high-performance network. To optimise the configuration of the BeeGFS storage nodes, several instances or targets were defined—one for storage and another for meta data that is required for the parallel access of the entire existing cluster computing node (nine computing nodes in total).
To maintain this structural investment, a three-year BeeGFS support contract was signed with Bechtle as a certified GOLD partner for the solution, as well as the extension of the support contract with Bright Computing.
Business benefits.
The HPC resources have been upgraded two-fold—through advanced software on the Bright Cluster Manager (upgrade to version 8.1) and with improved hardware thanks to the deployment of the new parallel BeeGFS storage solution.
The result is that all machines can write to and read the storage nodes without creating a bottleneck. What’s more, the architecture can be scaled for use into the future.